Abstract: The safety of existing bridges causes a universal concern of engineers and administrators. Traditionally, the vehicle load effect is determined according to the code in the absence of site informations. Recently, the installation of a WIM system enables a more accurate approach to estimate the vehicle load effect; however the “tail sensitivity” in a probalisitic model of vehicle load effects halters the application of WIM data. To overcome this problem, an improved GPD model is proposed in this paper to fit the tail distribution of the load effect based on the recorded WIM data. Firslty, the GPD formula was altered to fit the characteristics of vehile flow; then parameters of a GPD model was determined according to the Minimum Squre root of an Error (MSE) principle, and further modified for the case of “cut tail”. Using the proposed model, the on-site vehicle data were analyzed and the extreme load effect was estimated, and compared with the code result further, demonstrating that the proposed method is accurate in estimating extreme traffic load effects for a longer period, while the code method may underestimate the extreme value when applied to a medium period or shorter.